NBM V4.0 Now "Live"

Feb 24, 2021

NBM v. 4.0 Screenshot

Last fall, scientists from MDL’s Statistical Modeling Division (SMD), in concert with NCEP Central Operations staff, successfully implemented the newest version of the National Blend of Models (NBM) into regular production.  NBM v4.0 represents a significant leap forward in Blend capability, most notably in that it includes a significant expansion of the amount and types of probabilistic information available to its users.  Of course, effectively using and understanding this wealth of new probabilistic products and parameters can present its own set of challenges.   To this end, MDL’s related forecast visualization tools also have progressed alongside advancements in the NBM.  New capabilities recently introduced into the MDL Whole Story Uncertainty and Probabilities (WSUP) viewer promise to unlock a broad, new spectrum of possibilities in the visualization, packaging, and utilization of NBM probabilistic information.

The Blend now includes standard deviation information for temperature, dewpoint, maximum/minimum temperature (MaxT/MinT), surface-based convective available potential energy (SBCAPE), and significant wave heights over all domains.  Over the CONUS, the probabilistic MaxT/MinT information includes percentiles of the cumulative frequency distribution (5th, 10th, 25th, 50th, 75th, 90th, and 95th), in addition to exceedance probabilities for certain thresholds.  Additional percentiles also have been added to the previously-existing set for snow/ice accumulation (see thumbnail image and Fig. 1, below), wind speed, mean sea-level pressure, and precipitation amount (QPF).  New, calibrated probabilistic ceiling and visibility forecasts have been added to the suite of products for the CONUS as well.  For all domains except Guam, improved 1-, 3-, 6-, and 12-h probabilities of observed thunder are also available.  This new information supplements the more rudimentary probabilistic guidance that has been available in the Blend for some time.

Figure 1:  Composite of NBM v4.0 48-h accumulated snowfall amount percentiles from the 1/29/2021 1700 UTC run (81-h forecasts, valid 1200 UTC 2/02/2021).

In addition to the new gridded products available graphically in the WSUP viewer, users of the Blend’s NBH, NBS, NBE, NBX, and NBP station text bulletins will find that several of the above parameters are now available in this format as well.  (These can be found at: https://www.weather.gov/mdl/nbm_text.)  Also, we are now providing climatological MaxT/MinT, PoP12, and weekly-averaged QPF values on the NBM extended-range messages, where this information is available.  Over 2,000 new sites have been added with NBM v4.0, bringing the current total to about 9,000 and including most CONUS public airports.  We encourage longtime users of the similar Model Output Statistics (MOS) point bulletins to become familiar with the NBM text products; not only do they provide a wide range of additional meteorological parameters, but also are designed to leverage information from multiple NWP models.

A significant number of changes to NBM v4.0 methodology and products have been made to improve both probabilistic and deterministic guidance for aviation interests.  For example, the new probabilistic ceiling and visibility categories were chosen specifically to provide information about the likelihood of experiencing conditions below important operating criteria (MVFR, IFR, LIFR).  Changes also have been made which more directly leverage the most skillful Blend inputs for forecasts of these elements, substituting Localized Aviation MOS Program (LAMP) guidance for NBM projections of 1-36 hours, and utilizing probabilistic Global Forecast System-based MOS and North American Model- based MOS for forecast projections of 37-84h.

We also have made changes which improve NBM v4.0 guidance for METAR reporting sites by substituting observed METAR values for the NCEP Unrestricted Mesoscale Analysis (URMA) at the corresponding gridpoints for several weather elements.  This replacement allows all available input models for MaxT/MinT, temperature, dewpoint, wind speed, and wind gusts to be calibrated to METAR observations rather than the URMA analysis at those locations.  On the other hand, experience and verification have shown that the NCEP analysis is the superior ground truth for NBM sky cover guidance; we are now calibrating the NBM v4.0 sky cover forecasts to the URMA.  Together with the new ceiling and visibility information, these changes should further enhance the utility of NBM guidance for daily airport forecasts and operational planning.

To illustrate the effects of some of these changes, Fig. 2 contains an example NBM v4.0 wind speed plot and a 1-month verification of wind speeds (May, 2020) comparing NBM v4.0 wind performance to the earlier NBM v2.3 at approximately 2,300 METAR locations over the CONUS.  NBM v4.0 winds are stronger and more realistic, especially over mountainous terrain.  The figure also shows a considerable reduction in overall Mean Absolute Error (MAE).  Figure 3 is a comparison of sky cover fields for the two Blend versions at identical forecast valid times.  NBM v4.0 sky cover amounts show more realistic coverage as a result of the calibration to URMA, especially with offshore stratus along the Pacific coast.  

Figure 2:  (a) Example NBM forecast wind speeds over the CONUS (8-h forecast, valid 1100 UTC 1/27/2021), and (b) comparison of NBM v4.0 and v3.2 wind speed Mean Absolute Error (MAE) for 1-260h forecast projections (x axis), May 2020

Figure 3:  Comparison of NBM v4.0 and v3.2 30-h sky cover forecasts valid 4/21/20.

NBM v4.0 features a number of advancements in postprocessing methods, including expanded use of the Quantile Mapping and Dressing (QMD) technique developed in collaboration with scientists at NOAA’s Physical Sciences Laboratory.  The QMD technique has been used previously to generate calibrated probabilistic precipitation forecasts (PQPF) over the CONUS.  These have been further tuned to historical data in order to help improve Blend forecasts of heavier precipitation amounts.  This has resulted in improved skill compared to similar forecasts from NBM v3.2, which can be seen in the observed threat scores of Fig. 4 for the 0.5-in threshold.   New QMD PQPF guidance also has been added for the oceanic domain, which utilizes ground-truth datasets from the Climate Prediction Center’s Morphing Technique, or “CMORPH.”  Beyond the realm of precipitation forecasts, QMD is being extended to the generation of probabilistic temperature forecasts for the first time in NBM v4.0, including the new MaxT/MinT percentiles mentioned previously.

Figure 4:  Comparison of NBM v4.0 and v3.2 QPF threat scores by forecast projection for 6-h amounts greater than or equal to 0.5 in, May 2020.

NBM v4.0 methodology also includes the first use of retrospective runs to help improve the weighting of Blend inputs as part of its dynamic bias correction procedure, and now employs the square of the MAE in lieu of the absolute value for weighting of several CONUS weather elements (temperature, dewpoint, MaxT/MinT, wind speed, wind gusts, and sky cover).  As with most of the projects ongoing in MDL, the NBM is very much a work in progress.  We can look forward to many other enhancements to Blend information and statistical techniques in the months and years ahead, as well as to increased capability and flexibility in using them.